- 1 Development Logs
- 1.1 Project Details
- 1.2 Introduction
|Project Name||Benchmark Performance Database|
|Project Sudent||Ankesh Anand|
|Phone Number||+91 8348522098|
This page will contain weekly targets and Updates about the work done.
Community Bonding Period
(23rd April to 18th May)
- Got familiar with the modules and database schema from previous repository.
- Had a discussion with H.S. Rai(Project Mentor) about the infrastructure to be used. We settled on Django as a web framework for the project.
- Learnt Django by reading through books and building a couple of small applications with it.
- Set up the development repository on Github.
- Discussed about the deployment options on BRL-CAD servers. It already has Python and MySQL support, and since we manage it ourselves, it would be easy install new packages on it.
- Fixed a few bugs to get the database and parser portions fully working. The benchmark logs can now be parsed and indexed into the database from the command line.
- Built a Django app with jquery to support file uploads that let's users(Link to the code is here):
- Upload Benchmark logs
- Does validation checks
- allows the admin to view a list of files.
Week 1 (19th-25th May)
- 19th May(Monday): Started working on the plots app. Integrated the legacy database from previous repo with the models in the plots app.
- 20th May(Tuesday):' Continued the work with plots app, integrated both the databases(fileupload and benchmark) to a single database. Read the Django documentation on views and templates to refresh and understand a few details. Updated url mappings, and a simple admin interface for plots app.
- 21st May(Wednesday): I went through the tutorials on Highcharts and emulated a few examples. Created simple bar and line charts to understand the concept of series, axes and how the data is loaded via AJAX.
- 22nd May (Thursday): I decided to JSON as the data source for plots. This would require me to pass JSOn objects from view to templates. I went through the json module of python, particularly json.dumps and json.encode functions which will help in parsing JSON. I also thought about the types of plots I could generate from our app.
- 23rd May (Friday): Brainstormed over the visualizations I should work on, referred to suryajith's work on this and had a look at Openbenchmarking.org. After a discussion on the mailing list, we were able to arrive at a preliminary set of plots I should work on. Fixed a bug in the database connection file. Also got in touch with Sean regarding a server account.
- 24th May(Saturday): Half day of work. I created some charts using AJAX and a JSON file with initial data with the Highcharts library, had a few hiccups along the way. Next step is to refine these prototypes and integrate with the Django application.
Week 2 (26th May-1st June)
- 26th May(Monday): Did a comparison of various charting frameworks as Rai Sir had asked for. Also mentioned the reasons of choosing the HighCharts library. Further experimentation with HighCharts,I am able to load the series and chart options dynamically now. The url configurations have been modified to provide appropriate mapping for the charts and their data. Threw a couple of stub views which now I have to worked on. Cleaned up the fileupload app which was a mess of useless JS files.
- 27th May(Tuesday):
- Forked the repository to BRL-CAD's Github organisation, subsequent development will take place there. The new home for the repository is: https://github.com/BRL-CAD/benchmark
- Wrote the Django template file for charts. This contains the chart config and the JS script to make AJAX calls for JSON data. (https://github.com/BRL-CAD/benchmark/blob/master/plots/templates/chart.html)
- Started work on plots/data.py: The script which will return the aggregated data from database in the form of a dictionary. Wrote the aggregation codes for Avg VGR vs Processor Graphs, other plots should follow this structure. (https://github.com/BRL-CAD/benchmark/blob/master/plots/data.py)
- Wrote the rawdata view in plots/views.py that dumps JSON string after invoking the appropriate data aggregation function from data.py
- 28th May(Wednesday): Resolved a bug which prevented sending AJAX requests for the raw data in form of a JSON string in Django. Added support for two more plots, Average VGR Rating vs Number of Cores and Average VGR Rating vs OS Type. Made the chart template dynamic so that we don't have to change it depending on the plot. Added doc strings to views and data files.
- 29th May(Thursday): Added support for 4 more plots: Logarithmic VGR Rating vs Processor Family, Logarithmic VGR Rating vs OS Type, Logarithmic VGR Rating vs Number of Cores ,Running Time vs Processor Family. Spent some time resolving a bug in Django ORM.
- 30th May(Friday): Another plot supported now, the Absolute Rays per sec against Reference Images plot is up. Also added the options to export chart in form of a PDF, or an image. This is how one of the plot looks like:
- 31st May(Saturday): Had to hold off work on HighCharts due to licensing issue. Spent some time figuring out the details of alternative libraries. People on the list have asked me to do a spreadsheet comparing different libraries, and I got started on that.
- 2nd June(Monday): Created prototypes with Flot to get a better understanding of the API references. Read through Flot's documentation, and was able to emulate earlier plots with plastic data. Also got flot charts to works via AJAX data.
- 3rd June(Tuesday): Ported all the plots from HighCharts to Flot. The template and data modules were modified accordingly. Here is a sample chart in Flot:
- 4th June (Wednesday): Resolved a few bugs in data module, changed the format of JSON data to better integrate with Flot, made the code a bit more modular so that it is flexible across all the plots.
- 5th June (Thursday): Drew mockups of the plots page, and started working on the index page in the plots app.
- 6th June (Friday): Made an initial prototype of the plots page. The different charts are loaded via AJAX, so there are on page transitions. Updated URL patterns, added custom styles in the process. This is a screenshot of the initial version, for a higher resolution, see here
- 7th June (Saturday): Fixed bugs in the front-end, added support for a new plot (Average VGR vs Number of Processors)
- 8th June (Sunday): Took the day off.
- 9th June (Monday): Read a few articles on the science behind data-visulaizations and how to achieve better and meaningful results. Also started working on multi-series plots and refereed to the documentation accordingly.
- 10th June (Tuesday): Worked on integrating both the plots and upload applications. Also ensured that they are decoupled so that the code remains independent and modular, besides this is also a recommended Django design principle. Related UI fixes and configuration changes were carried out in the process.
- 11th June (Wednesday): Added efficiency plots(Average VGR Rating against CPU MHz and no. of cores). Studied open benchmark suites and looked at openbenchmarking.org to get a better understanding of the end-usage aspects of a benchmarking tool. Also, requested Sean to provide logs which will assist in testing.
- 12th June (Thursday): Completed the integration of the front-end interfaces of plots and the upload app. Also ensured the line charts and their labels are automatically scaled when the data is not discrete. I still await the benchmark logs from Sean, but I will move forward next to creating scripts that check for new benchmark logs.
- 13th June (Friday): Leveraged the earlier script written by Surjyajith to process files from the queue. Made relevant changes to get it working. The scripts needs to be run periodically, we can use crontab for that. I enquired on the IRC channel and Erik confirmed that crontab is installed on the servers.
- 15th June (Sunday): Added the multi-series plot Performance of Processor Families against Reference Images. Along with that added two plots in the Running Time Section. Fixed some issues with line chart, and added comments to the index page. With this, I think the initial roaster of plots is complete, however I still need to find a way to work with Logarithmic axis for Log charts and think about Distribution charts.
- 16th June (Monday): Caught a minor flu so couldn't accomplish much work today. Had a relook on the documentation and tried to brain-storm further aspects of the project. Tried reaching Sean again unsuccessfully, I need further perspectives on the project I think, will draft a detailed mail on the list if IRC discussions are not possible
- 17th June (Tuesday): Tried to figured out the problems with logarithmic axis in Flot, it turns out logarithmic axis are not supported out of the box and we need to provide ticks manually. Thought of solutions to serve the performance data to developers, and took notes from OpenBenchmarking.org in this aspect. This week is reserved for documentation and testing, so I asked on the list if I should continue the documentation now. Also posted some issues and questions I had.
- 18th June (Wednesday): Had a great IRC discussion with Sean about several aspects of the project. We discussed the possible views of the application, and the importance of displaying immediate results to the user, alongside comparisons with the database. Regarding this I look into the speedtest portal to get an idea of the UI aspects. I will hold off on docs now as advised.
- 19th June (Thursday) and 20th June(Friday):
- Couldn't update the logs yesterday due to sudden internet outage. I have worked on making the website consistent with the BRL-CAD color scheme. I posted a screenshot on IRC.
- Apart from this I worked on making the immediate results view of the uploaded logs. I explored how to resolve the issue of conflicting file names. Discovered that Python has a uuid library which produces unique strings, which can be appended to the filename to get a unique name. To generate unique URLs, I can use the hash of this filename and serve the results on that URL to the user.
- 21st June(Saturday): Started creating the module for showing individual results. Prepared mock-ups of the design after having a look at a few dashboard sites and other relevant links. Also, fixed a few issues in the new color-scheme.
- 23rd June (Monday): Continued the work on results module. Create the starting models and views. Since the models used by this module and the polls module would be same, I looked for ways to share them, and found an answer from the Django community. I worked on UI as well and started implementing my mock-up.
- 24th June (Tuesday): Studied various dashboards to derive inputs from them, continued to work on the frontend of the results page, implemented demo scripts for hashing the file-name and then generating a unique URL as well. For now the unique URL length is rather too long, I will study further to see how I can compress it.
- 25th June (Wednesday): Did a bit of code-cleanup and some UI fixes. Also started on the blog for midterm evaluations, a half-draft is done and plan to complete the rest by tomorrow.
- Midterm Report: http://www.ankeshanand.com/posts/gsoc-midterm-status-quo/
- 26th June (Thursday): Finished writing off the blog today and studied about writing Python setup scripts from here: https://docs.python.org/2/distutils/setupscript.html
- 27th June (Friday): Read about working with Virtual Environments in Python. Deployed the application at: http://184.108.40.206:3000
- 28th June (Saturday): Worked on a sample application using django social-auth for managing user accounts and integrating logins from OAuth providers. Tried to resolve a file upload bug un-successfully even after logging stdout and stderr as there were no error messages logged. Will try to resume it tomorrow.
- 30th June (Monday):
- Ensured unique filenames for all uploaded logs using Python UUID4.
- Modified models to make sure all are accessed by Django ORM.
- Added show_result view to immediately parse the log and it's information is inserted into the database. The data for the file is then extracted using Django ORM and returned as a dictionary to the result template.
- 1st July (Tuesday): Completed the template markup, linked the view to the template. The front-end now shows all the information extracted from the file immediately uploaded. Next step is to trigger AJAX request from the fileupload app to the results page.
- 2nd July (Wednesday): Was sick due to a minor fever and headaches. Took medication and hope to compensate in the upcoming days.
- 3rd July (Thursday): Completed the Results view I was working on. Here's a screenshot, for a larger version, click here. Next up, I have to integrate this with the file-upload app, so that results are shown automatically on file-uploads.
- 4th July (Friday): Encountered a bug while integrating the fileupload and results module. The database changes were not reflected in Django ORM due to query caching. Tried flushing the transaction, but it didn't work, will look more into this.
- 6th July (Sunday): Integrated the results module with fileupload. Now, upon uploading benchmark logs, users are shown the Benchmark Result immediately. Resolved the bug I was facing earlier my doing a configuration change in MySQL. Also, ensured database connections are closed after parsing the file, which was not the case earlier. Tried deploying the new code to the remote machine, but couldn't login through ssh, maybe the system was shut down.
- 7th July (Monday): Couldn't get much work done due to a faulty internet connection.
- 8th July (Tuesday): Deployed the application on the server. Resolved a bug related to CSRF tokens which caused file-uploads to stop midway. Discussed further project plans with Sean on IRC.
- 9th July (Wednesday): Explored the feature of uploading files via a web POST API. I found a library django-tastypie that solves the issue of easily creating APIs. Went through it's documentation and tutorials to understand it better.
- 10th July (Thursday): Worked on an example of POST API with django-tastypie. Was able to do fileuploads with an API key on the example. Also looked at Geekbench Benchmarks to get an understanding of different benchmark systems.
- 11th July (Friday): Went through the documentation on Django Forms in detail. The comparisons page would require filtering from the database, and hence Django Forms would be useful there.
- 14th July (Monday) to 16th July(Wednesday): My institute opened from Tuesday, so I had to pack my luggage and travel from my home to my institute on Monday. It took a day to complete the registration and fee payment there, and another day to fix logistics as I had to shift to a new accommodation.
- 17th July (Thursday): Back to work resuming the code for the Comparisons module. Wrote it's further backend logic based on Django Forms ORM which would help to communicate with the database based on different filters. I will set up the demo back tomorrow and continue further work on the module.
- 18th July(Friday): Completed the backend logic of the Comparisons view. After selecting the options, a POST request is now sent which returns the appropriate filtered data. The user is shown his VGR Rating and the Average VGR Rating among the filtered logs. Next up, I have to present the data in a better manner, and include a graph for comparison.
- 21nd July (Monday): Almost done with the comparisons module, have added the filters and completed backend querying code. Here's a screenshot, for a larger version, click here.
- 22rd July (Tuesday): Added a Date filter to the comparisons view. It lets filter down logs submitted this week, this month, this year or at all times. Added More-Info modal dialogs which give detailed information of the ratings in the result and comparison view.
- 23th July (Wednesday): Completed minor changes in the Comparisons module and started working on the POST API for file uploads. I am using django-tastypie which facilitates easy development of apis for django projects. Will host the project on BRLCAD's server once this module is done, which would be 2-3 days probably.
- 24th July (Thursday): Continued work on the Web API for file uploads. Was able to send get requests via the API but POST requests were unsuccessful. Also, fixed a few minor bugs and redeployed the demo.
- 26th July (Saturday): Worked on deploying the app to production. Read lots of documentation and articles on this to understand every bits and pieces. Set up Post Receive hooks in the remote repository so that it is synced with Git. Created a makefile for easy installation of dependencies and carry out other deployment related tasks. Will contact admins from the IRC channel tomorrow to set up the apache mod_wsgi configurations and get everything running.
- 27th July (Sunday): Did a bunch of changes to get the production and development environment in proper setup and running. New installation instructions are now in place. Today's commits can be accessed here: https://github.com/BRL-CAD/benchmark/commits/master.
- 29th July (Tuesday): Completed the Web API for file uploads. Now, files can be uploaded from the command line via a POST request such as:
- curl -F "slug=test" -F "firstname.lastname@example.org" http://localhost:8200/upload/api/v1/picture/
- 30th July (Wednesday): In some cases the files uploaded via the Web api were being partially uploaded. Added a deserializer method to take care about of that. Also refractored the api code.
- 31st July (Thursday): Refactored some of the code, and did some mockups of Live feed for the benchmark home page.
- 1st August (Friday): Looked into automatic uploading of logs from the benchmark script. I am able to make file uploads from the script, but need to find a way to make it fault tolerant. I tried a few suggestions mentioned on SO but they were not completely solving the issue. Will dig into this further.
- 3rd August (Sunday): Checking off things from the TODO. Made the social media sharing icons work. Fixed some of the broken links in the navbar. Started working on the home page and file archiving, had a few doubts regarding them, so sent a mail to the mailing list.
- 4th August (Monday): Started work on the homepage created a few mockups and continued code-refactoring. Awaiting response from the mailing list about my questions.
- 5th August (Tuesday): Found the solution to the file archiving issue. The recommended way is to organize files in yy/mm/dd directories so that they are searchable by date and less number of files in a directory would also significantly improve performance. Started implementing the home page mockup, refactored some code in the results page.
- 6th August (Wednesday): After experimenting with different designs, settled down on the landing page. I have implemented the mockup, below is a screenshot.
- 7th August (Thursday): Pushed the site to deployment after enabling the site in apache, it is now live again at http://220.127.116.11/benchmark/. Tested the site across multiple platforms, resolved most of the UI fixes. There is a top margin issue that's left to be resolved. Will work on the file archiving issue tomorrow.
- 8th August (Friday): Implemented file-archiving, the files are now segmented across separate directories for each month. This ensures they are easily searchable and also significantly improves the speed of the app.
- 11th August (Monday): Wrote the structure for application feed. The feed would list out all the recently uploaded log files with links to results page for each. I have to plug in the ORM statements to make it working.
- 12th August (Tuesday): Modified the production setup to use virtualenvs as this was causing the other projects to come down. This included a different make configuration and changes to the WSGI file. The setup was also moved from /var/www to /home/benchmark/deployment. The aliases were configured accordingly
- 13th August (Wednesday): Wrote a view that lists out the most recently uploaded logs by listening to the database for newly uploaded logs. This can be merged with the homepage later.
- 14th August (Thursday): Worked on the script to extract logs submitted via email. Leveraging some part of the work that was done by Suryajith. The IMAP library has limited support now but the script works, tried to port this into the new library by GMail.
- 15th August (Friday): Completed the port of the script to extract email logs. Improved parser scripts to remove hardcoded variables, and use project settings instead.
- 16th August (Saturday): Improved inline documentation in the results and fileupload module. Resolved the issue of navbar responsiveness on iPad. Few other UI tweaks along the way.
- 18th August (Monday): My lab had a visit from Prof. Khosla, the Chancellor of UCSD. I was busy preparing a presentation regarding the same and then giving a talk on the current ongoing research in the lab.
- 19th August (Tuesday): Wrapped up everything and submitted the final evaluation. The whole experience was great and I can definitely say GSoC was one of the best things I have ever done. Thanks to Prof. Rai and Sean for being so helpful along the way.